Abstract
Computer diagnosis of electrocardiogram (ECG) has been extensively studied in this country, but is not yet widely used by general practitioners. A computer programm was worked out to process a standard ECG, recorded with a popular direct-writing cardiograph. Input data for this program were prepared from measurements of 15 parameters and answers to 65 questions according to a questionnaire. This questionnaire can be filled by an ECG technician after 30 minutes' training. It takes 6 minutes on an average to fill each questionnaire.
The validity of the program was tested by samples from both textbook data (261 cases) and hospital records (335 cases). 47% of diagnoses given for textbook data proved satisfactorily consistent with human diagnosis and 25% only partly and tolerably consistent. The rest (29%) was found to be inconsistent. 75% of diagnoses given for hospital records proved satisfactorily consistent with human diagnosis, 15% only partly and tolerably consistent, and the rest (7%) inconsistent.
On the other hand, almost all ECGs (4968 cases) recorded at the Kambara General Hospital (KGH) for the past three years were statistically analysed regarding their diagnoses. Combining this result with that of textbook data analysis, at least about 70% of KGH data can be expected to be fully or tolerably analysed by the computer. Simplification of the questionnaire format was one reason for the low efficiency of arrhythmia processing ; it limited the detection of comparatively rare abnormalities. Moreover, this was not very effective in decreasing measurement errors and inter-observer variations.
Despite these shortcomings, ECG diagnostic service by this program might be valuable in its capability to process popular ECG records and in its inexpensiveness.
